NP-completeness. Chapter 34. Sergey Bereg

Transcription

1 NP-completeness Chapter 34 Sergey Bereg Oct 2017

2 Examples Some problems admit polynomial time algorithms, i.e. O(n k ) running time where n is the input size. We will study a class of NP-complete problems which will be defined later. No polynomial time algorithm is known for any NP-complete problem. Shortest vs. longest simple paths. Given a directed graph G = (V, E) with non-negative edge weights, the shortest path from between two vertices can be found in O(V E) time. Finding the longest simple path is a NPcomplete problem. What if we modify Bellman-Ford algorithm by changing > to < in the relaxation condition? Relax(u, v, w) // relax edge (u, v) using weight w(u, v) 1 if d[v] > d[u] + w(u, v) then 2 d[v] = d[u] + w(u, v) 3 π[v] = u Oct

3 Examples Some problems admit polynomial time algorithms, i.e. O(n k ) running time where n is the input size. We will study a class of NP-complete problems which will be defined later. No polynomial time algorithm is known for any NP-complete problem. Shortest vs. longest simple paths. Given a directed graph G = (V, E) with non-negative edge weights, the shortest path from between two vertices can be found in O(V E) time. Finding the longest simple path is a NPcomplete problem. There is a graph G and a vertex s G such that a modified Bellman-Ford algorithm does not find the longest simple paths. Oct

4 Definitions An abstract problem Q is a binary relation on a set I of problem instances and a set S of problem solutions. Example. An instance for Shortest-Path is a graph G = (V, E) and two vertices. A solution is an ordered sequence of vertices (can be empty sequence if no path exists). A given instance may have more than one solution. A decision problem is a problem having yes/no solution. So S = {0, 1}. Many problems are optimization problems where some value must be minimized or maximized. Usually it is easy to cast an optimization problem as a decision problem. Example. Decision problem for Shortest-Path is the problem Path: Given a graph G = (V, E), vertices u and v, and an integer k, is there a path of from u to v consisting at most k edges? Oct

5 Encodings Encoding of a set S is a mapping e : S {0, 1} from S to the set of binary strings. Example: S = N, then binary code e : N {0, 1}. A problem whose instance set is {0, 1} is called a concrete problem. We say that an algorithm solves a concrete problem in time O(T (n)) if it takes O(T (n)) time for a problem instance i of length n = i. If T (n) = O(n k ) then we say that the problem is polynomially solvable. Complexity class P is a class of concrete decision problems that are polynomially solvable. Oct

6 Formal language framework An alphabet Σ is a finite set of symbols. A language L over Σ is any set of strings made up of symbols from Σ. Example: if Σ = {0, 1}, the set L = {10, 11, 101, 111,... } is the language of binary representations of prime numbers. We denote the empty string by ε and the empty language by. So, L Σ where Σ is the language of all strings on Σ. Operations on languages: union L 1 L 2, intersection L 1 L 2, complement of L is Σ L, concatenation of L 1 and L 2 is {x 1 x 2 : x 1 L 1, x 2 L 2 }, closure or Kleene star L = {ε} L L 2 L 3... where L k is the language obtained by concatenating L to itself k times. Oct

7 Formal language framework An algorithm A accepts a string x {0, 1} algorithm s output is 1, i.e. A(x) = 1. A rejects x if A(x) = 0. if, given input x, the The language accepted by an algorithm A is the set of strings L = {x {0, 1} : A(x) = 1}. Question: Is it true that A rejects a string x / L? Oct

8 Formal language framework An algorithm A accepts a string x {0, 1} algorithm s output is 1, i.e. A(x) = 1. A rejects x if A(x) = 0. if, given input x, the The language accepted by an algorithm A is the set of strings L = {x {0, 1} : A(x) = 1}. Question: Is it true that A rejects a string x / L? No, A may not reject a string x / L (A may loop forever on x). Oct

9 Formal language framework A language L is decided by an algorithm A if A accepts/rejects every string x in/not in L. A language L is accepted in polynomial time by an algorithm A if, for any x L, x = n, it accepts x L in O(n k ) time for some fixed k. A language L is decided in polynomial time by an algorithm A if, for any x {0, 1}, x = n, it decides whether x L in O(n k ) time for some fixed k. P = {L {0, 1} : A that decides L in polynomial time}. Theorem. P = {L {0, 1} : A that accepts L in polynomial time}. Oct

10 Formal language framework Theorem. P = {L {0, 1} : A that accepts L in polynomial time}. Proof. If a language L is decided by a polynomial-time algorithm A, then L is accepted by A. Suppose that a language L is accepted by a polynomial-time algorithm A with running time at most T (n) = cn k. We use a simulation argument and create a new algorithm A. For an input string x, the algorithm A 1. Computes the length n = x. 2. Simulates the action of A only for time T (n). If A accept/rejects x then A reports it and stops. 3. Rejects x. Oct

11 Verification algorithms Hamiltonian cycle problem HAM-CYCLE. Given an undirected graph G, is there a simple cycle that contains all the vertices of G? An algorithm: encode graph with adjacency matrix, the number of vertices m n. There are m! permutations and the running time to check them is at least Ω(m!) = Ω(( n)!) = Ω(2 n ). Not polynomial time. To verify a solution of HAM-CYCLE we need a certificate, for example, the hamiltonian cycle. A verification algorithm has two arguments - input string and certificate string. A two-argument algorithm A verifies an input string x if there is a certificate y such that A(x, y) = 1. The language verified by a verification algorithm A is L = {x {0, 1} : y {0, 1} such that A(x, y) = 1}. Oct

12 Hamiltonian Game Is there a walk along a closed path on dodecahedron such that every vertex is visited one time? Oct

13 Complexity class NP Complexity class NP is the class of languages that can be verified by a polynomial time algorithm. So, L NP if and only if there exist a two-input polynomial-time algorithm A and a constant c such that L = {x {0, 1} : y {0, 1} such that y = O( x c ) and A(x, y) = 1}. We say that algorithm A verifies L in polynomial time. Example: HAM-CYCLE NP. If L P then L NP since there is a polynomial-time algorithm to decide L (the verification algorithm ignores the second argument). Thus, P NP. Whether P = NP is the famous problem. It is not even known if NP is closed under complement. That is, does L NP imply L NP? Complexity class co-np is the set of languages L such that L NP. So, the question is NP=co-NP? Oct

14 Classes P, NP, and co-np Since P is closed under complement, we have P NP co-np. There are 4 possible relations between P, NP and co-np. P=NP=co-NP NP=co-NP P co-np P NP co-np NP P Oct

15 Reducibility A language L 1 is polynomial-time reducible to a language L 2, written L 1 P L 2, if there exists a polynomial-time computable function f : {0, 1} {0, 1} such that for all x {0, 1} x L 1 if and only if f(x) L 2. f is the reduction function and a polynomial-time algorithm F that computes f is called a reduction algorithm. Example. Equal-Numbers P Same-Points. Equal-Numbers: Given n integer numbers, are there two equal numbers? Same-Points: Given n points in the plane with integer coordinates, are there two coinciding points? Oct

16 Reducibility Lemma. If L 1, L 2 {0, 1} are languages such that L 1 P L 2 and L 2 P, then L 1 P. Oct

17 Reducibility Lemma. If L 1, L 2 {0, 1} are languages such that L 1 P L 2 and L 2 P, then L 1 P. Proof. Consider the algorithm for L 1 that is combination of the reduction algorithm and the algorithm for L 2. Oct

18 NP-completeness A language L is NP-complete if 1. L NP, and 2. L P L for every L NP. We say that a language is NP-hard if it satisfies condition 2 (1 may be satisfied or not). Any NP-complete problem is NP-hard. Example of NP-hard problem that is not NP-complete (does not satisfy the condition 1): halting problem (given a program and its input, will it run forever?). Theorem. If any NP-complete problem is solvable in polynomial time, then P = NP. Equivalently, if some problem in NP is not solvable in polynomial time, then no NP-complete problem is polynomial-time solvable. Oct

19 Classes P, NPC NP NPC P How most theoretical computer scientists view the relationships among P, NP, and NPC. Oct

20 Circuit Satisfiability A boolean combinatorial circuit is composed of AND, OR, and NOT gates connected by wires. Circuit input is a wire that is not the output of a gate. Circuit output is a wire that is not the input of a gate; it is unique. A truth assignment of a circuit is a set of boolean input values. x z x y z x y z x x x y x y x y x y A boolean combinatorial circuit is satisfiable if it has a truth assignment that causes the output to be 1. This assignment is called a satisfying assignment. Oct

21 Circuit Satisfiability Example. x 1 = 1, x 2 = 1, x 3 = 0 is a satisfying assignment and the circuit is satisfiable. x 1 = 1 x 2 = x 3 = Oct

22 Circuit Satisfiability Circuit-SAT problem. satisfiable? Given a boolean combinatorial circuit, is it Lemma. Circuit-SAT is in NP. Proof. Make a verification algorithm A(x, y) where x is an encoding of circuit and y is the certificate (satisfying assignment). A must be a polynomial-time algorithm. Idea: for each logic gate A computes its output based on inputs. At any moment, there is a gate with ready inputs. A returns the final output (of entire circuit). Oct

23 Circuit Satisfiability Lemma. Circuit-SAT is NP-hard. Oct

24 Circuit Satisfiability Lemma. Circuit-SAT is NP-hard. Proof. Let L be a language in NP and let A be a verification algorithm for A. We want to design a polynomial-time algorithm that maps every x {0, 1} to a circuit C = f(x) such that x L if and only if C Circuit-SAT. Idea: model the execution of the algorithm (code for A, program counter, machine state, and working storage) with a circuit C. There are T (n) = O(n k ) steps in the execution. We put T (n) copies of the circuit M that implements the computer hardware. We run the algorithm on the certificate y. Oct

25 Boolean formula satisfiability An instance of Circuit-SAT can be transformed into a boolean formula φ composed of 1. n boolean variables x 1,..., x n 2. m boolean connectives: any boolean function with one or two inputs and one output, such as (AND), (OR), (NOT), (implication. like ), (if and only if. like =) 3. parentheses x y x y x y Example: φ = ((x 1 x 2 ) (( x 1 x 3 ) x 4 )) x 2. Oct

26 Boolean formula satisfiability A truth assignment for a boolean function φ is a set of values of variables of φ and a satisfying assignment is a truth assignment that causes it to evaluate to 1. If a satisfying assignment exists then φ is satisfiable. SAT problem. Given a boolean formula, is it satisfiable? Theorem. SAT is NP-complete. Proof. 1. We show that SAT is in NP: certificate is the truth assignment that satisfies the formula. Can be checked in polynomial time. Oct

27 Boolean formula satisfiability 2. We show that Circuit-SAT P SAT. Assign a variable for each wire. Formula for SAT is the AND of the circuit output and the conjunction of clauses describing the operation of each gate. The formula can be produced in polynomial time. It is satisfiable if and only if the circuit is satisfiable. x 1 x 2 x 3 x 4 x 5 x 6 x 7 x 8 x 9 x 10 φ = x 10 (x 4 x 3 ) (x 5 (x 1 x 2 )) (x 6 x 4 ) (x 7 (x 1 x 2 x 4 )) (x 8 (x 5 x 6 )) (x 9 (x 6 x 7 )) (x 10 (x 7 x 8 x 9 )) Oct

28 3-CNF A literal in a boolean formula is a variable or its negation. A boolean formula is in conjunctive normal form, or CNF, if it is expressed as an AND of clauses, each of which is the OR of one or more literals. A boolean formula is in 3-conjunctive normal form, or 3-CNF, if each clause has exactly 3 distinct literals. Example: (x 1 x 2 x 3 ) (x 3 x 2 x 4 ) ( x 1 x 3 x 4 ). 3-CNF-SAT, or 3-SAT, asks whether a given 3-CNF is satisfiable. Theorem. 3-SAT is NP-complete. Proof. 3-SAT NP is similar to the proof that SAT NP (every instance of 3-SAT is an instance of SAT). Oct

29 3-CNF We prove that SAT P 3-SAT. Let φ be an input formula for SAT, for example φ = ((x 1 x 2 ) (( x 1 x 3 ) x 4 )) x 2. Construct a binary parse tree. Introduce new variables y i. y 1 y 2 y 3 y 4 x 2 x 1 x 2 y 6 y 5 x 1 x 3 x 4 Oct

30 Transform the tree into a formula φ : 3-CNF y 2 y 1 y 3 y 4 x 1 x 2 y 6 y 5 x 2 x 1 x 3 x 4 φ = y 1 (y 1 (y 2 x 2 )) (y 2 (y 3 y 4 )) (y 3 (x 1 x 2 )) (y 4 y 5 ) (y 5 (y 6 x 4 )) (y 6 ( x 1 x 3 )) Oct

31 Make truth table for each clause φ i. 3-CNF Convert φ i using 0 s in the table. For example φ 1 = y 1 (y 2 x 2 ) y 1 y 2 x 2 y 1 (y 2 x 2 ) Then φ 1 = (y 1 y 2 x 2 ) (y 1 y 2 x 2 ) (y 1 y 2 x 2 ) ( y 1 y 2 x 2 ) Oct

32 3-CNF φ 1 = (y 1 y 2 x 2 ) (y 1 y 2 x 2 ) (y 1 y 2 x 2 ) ( y 1 y 2 x 2 ) Applying DeMorgan s laws φ 1 = ( y 1 y 2 x 2 ) ( y 1 y 2 x 2 ) ( y 1 y 2 x 2 ) (y 1 y 2 x 2 ) This way we create a CNF formula φ so that each clause has at most 3 literals. We change those with 1 and 2 literals. Oct

33 NP-complete problems A clique in an undirected graph G = (V, E) is a subset V V of vertices, each pair of which is connected by an edge in E. In other words, a clique is a complete subgraph of G. The size of a clique is the number of its vertices. Clique Problem is to find a clique of maximum size in G. Decision problem: CLIQUE= { G, k : G is a graph with a clique of size k}. Naive algorithm needs Ω(k 2( ) V k ) time. Theorem. The clique problem is NP-complete. Proof. CLIQUE NP if we choose the clique as the certificate. CLIQUE is NP-hard if we show 3-SAT P CLIQUE. Oct

34 Clique Problem The reduction algorithm begins with an instance of 3-SAT problem φ = C 1 C 2 C k. We construct a graph G(V, E). For each clause C r = (l1 r l2 r l3), r we place 3 vertices v1, r v2, r v3. r We put an edge between vi r and vs j if r s and literals of vr i and vs j are not contradictory. Example: φ = (x 1 x 2 x 3 ) ( x 1 x 2 x 3 ) (x 1 x 2 x 3 ). C 1 x 1 x 2 x 3 C C 3 2 x 1 x 1 x 2 x 2 x 3 x 3 Oct

35 Clique Problem We show that the transformation of φ is a reduction. Suppose that φ has a satisfying assignment. Then each clause C r contains at least one literal li r assigned to 1. These vertices form a clique of size k in G. Conversely, suppose that G has a clique V of size k. No edges in G connect vertices in the same triple, so V contains exactly one vertex per triple. We assign 1 to each literal in the clique (we do not assign 1 to a literal and its complement). We assign any value to a variable whose literals are not in clique. All clauses are satisfied and φ is satisfied. Oct

36 Vertex-Cover Problem A vertex cover of a graph G = (V, E) is a subset V V of vertices such that, if (u, v) E, then u V or v V (or both). The vertex-cover problem is to find a vertex cover of minimum size in G. Decision problem: VERTEX-COVER= { G, k : graph G has a vertex cover of size k}. Theorem. The vertex-cover problem is NP-complete. Proof. VERTEX-COVER NP if we choose the vertex cover as the certificate. VERTEX-COVER is NP-hard if we show CLIQUE P VERTEX- COVER. Oct

37 Vertex-Cover Problem We define the complement of G as G = (V, E) where E = {(u, v) : u, v V, u v and (u, v) / E}. An instance of G, k of the clique problem reduces to an instance G, V k of the vertex-cover problem. Indeed, a set V V, V = k is the clique of G if and only if V V is a vertex cover of G. The reduction can be done in polynomial time. a b a b e f e f d c d c G, clique {a, b, c, d} G, vertex cover {e, f} Oct

38 Vertex-Cover Problem Approx-Vertex-Cover(G) 1 C = 2 E = E[G] 3 while E 4 let (u, v) be any edge of E 5 C = C {u, v} 6 remove from E every edge incident on either u or v 7 return C Theorem. Approx-Vertex-Cover computes a 2-approximation in O(V + E) time. Oct

39 Hamiltonian-cycle Problem A hamiltonian cycle in a graph G = (V, E) is a simple cycle that contains all vertices of G. G is hamiltonian if it contains a hamiltonian cycle. Hamiltonian-cycle problem is to detect if a graph G is hamiltonian. HAM-CYCLE= { G : G is a hamiltonian graph }. Theorem. The hamiltonian-cycle problem is NP-complete. Proof. HAM-CYCLE NP if we choose the hamiltonian cycle as the certificate. HAM-CYCLE is NP-hard if we show VERTEX-COVER P HAM- CYCLE. Use widgets and selector vertices to transform G to G. Polynomial time. G has a vertex cover V if and only if G has a hamiltonian cycle. Oct

40 Traveling-salesman problem Let G be a complete graph with integer cost c(i, j) to travel from city i to j. Traveling-salesman problem is to find a minimum-cost tour (hamiltonian cycle). 3 a d b 2 c traveling-salesman tour acbda of costs 7. TSP= { G, c, k : G is a complete graph and has a traveling-salesman tour of cost at most k}. Theorem. The traveling-salesman problem is NP-complete. Proof. TSP NP if we choose the tour as the certificate. Hardness: HAM-CYCLE P TSP. Oct

41 Subset-Sum Problem Given a finite set S N (natural numbers or positive integers) and a target t, the subset-sum problem is to find a subset S S whose elements sum to t. Example: S = {1, 3, 3, 9, 18}, t = 10. Then S = {1, 9} is a solution. SUBSET-SUM= { S, t : there exists a subset S S such that t = s S s}. Theorem. The subset-sum problem is NP-complete. Proof. SUBSET-SUM NP if we choose S as the certificate. Hardness: 3-SAT P SUBSET-SUM. Oct

42 Subset-Sum Problem Reduction. φ = C 1 C 2 C 3 C 4 where C 1 = (x 1 x 2 x 3 ), C 2 = ( x 1 x 2 x 3 ), C 3 = ( x 1 x 2 x 3 ), and C 4 = (x 1 x 2 x 3 ). x 1 x 2 x 3 C 1 C 2 C 3 C 4 v 1 = v 1 = v 2 = v 2 = v 3 = v 3 = s 1 = s 1 = s 2 = s 2 = s 3 = s 3 = s 4 = s 4 = t = Oct